Multivariate regression modelling analysis of mood swings in music therapy and its effect on the regulation of psychological states

Abstract

This study focuses on the quantitative evaluation and intervention optimization of emotional fluctuations in music therapy. By integrating biometric technology and multivariate regression modeling methods, a dynamic emotional analysis framework is proposed and its clinical translational value is verified. Based on multimodal physiological data (EEG, ECG, GSR) from 120 subjects (60 patients with depression/anxiety and 60 healthy controls) and real-time PANAS emotion scores, a time series regression model was constructed to achieve high-precision prediction of emotional states (RMSE=0.85 ± 0.10, R ²=0.77 ± 0.05). Validation showed that the model identified HRV low-frequency power (β=-0.41, p=0.003) and EEG beta wave energy (β=0.38, p=0.007) as key biomarkers, revealing that patients with depression had significantly higher HRV regulation efficiency for low-frequency music (BPM=60-80) than the healthy population (28.7% vs. 6.5%, p=0.017), while anxiety patients had a 41.5% decrease in skin conductance response density under high-frequency music intervention (p=0.003). Further research has been conducted to develop a lightweight model integrated with the Apple Watch (with a parameter size of 1.2MB and power consumption of 2.3W), and ISO 13485 certification has been completed within the framework of digital therapy compliance (with a pass rate of 78.3%), providing a quantifiable technical path and clinical level toolchain for personalized music therapy in the field of mental health.